[4 pages — 3,073 words]


To protect against a network failure while using Industrial Ethernet, users are seeking cabling topologies that remain functional under a single cable loss. There are four popular redundancy schemes for Ethernet: Link Aggregation (Trunking), Proprietary Ring, Spanning Tree Protocol (STP), and Rapid Spanning Tree Protocol (RSTP). Each of these approaches has a set of benefits and tradeoffs, however, the industry seems to focus only on one aspect of redundancy and that is performance. How quickly does the network repair itself? This recovery time depends upon the industrial protocols being used. Protocols such as Modbus/TCP and EtherNet/IP™ rely upon the TCP/IP suite of transport-layer protocols, and they play a major role in how a network recovers from a single cable fault. A test was conducted using the various redundancy schemes to determine typical recovery times and the results were tabulated.

Topology Options

When we discuss Ethernet redundancy schemes, we are talking about cable redundancy. How can an Ethernet network continue to function after a single cable failure? This implies there is an alternate route to carry network traffic when a failure occurs in the primary path. However, Ethernet's star topology is not conducive to cable redundancy.

Star, Distributed Star and Tree Topologies

With star topology, stations are interconnected using a wiring hub. By connecting a wiring hub to another wiring hub, we introduce the distributed star or tree topology. Wiring hubs are available as either repeating hubs or switching hubs. With star topology, end stations connect to individual ports on a wiring hub. This type of wiring is convenient to implement in a plant but still does not provide for wiring redundancy. However, the development of switched Ethernet technology can provide a solution to the cable redundancy problem.

Switched Ethernet Technology

Switching hubs offer several advantages over repeating hubs. Unlike repeating hubs that function at the physical layer of the ISO Reference Model, switching hubs operate at the data link layer. The IEEE calls this class of equipment bridges, but they are commonly referred to as switches. Unlike repeating hubs, switches store-and-forward complete Ethernet frames. Switches segment the Ethernet network into separate collision domains. This allows for virtually unlimited geographic expansion of a network by the simple cascading of additional switches. Switches improve network throughput by limiting traffic to only those segments that are party to the communication. Switches learn the location of end stations by observing the source address within Ethernet frames that traverse the switch. The switch takes note of the station address-port number relationship in its address table. IEEE would call the address table the filtering database and the process of creating assignments would be called learning. Subsequent transmissions to a learned station would only be directed to its assigned port. This process is called forwarding. If the switch does not know the exact location of a station, it would forward the frame to all switch ports. This is called flooding. Since it is possible to physically move cables from one port to another, the address table could become invalid preventing a station from receiving its messages. To correct for this, address table assignments are periodically cleared forcing the relearning of the station port-assignments. This process is called aging.

Trunking or Link Aggregation

This redundancy approach maintains the star topology but instead of having one path between switches, two or more parallel paths are used. These multiple paths are called a trunk group and function as one larger channel. Complete frames are alternately sent down each of the parallel paths and recombined at the other end. By using multiple paths, the throughput increases with the number of separate paths. In some instances, a failed path will result in the data being diverted over the functioning paths, providing cable redundancy. This method is called "trunking." The IEEE has standardized this approach as Link Aggregation, but not all switches support this feature.

The advantage of trunking is that is provides an incremental increase in throughput as parallel paths are added. Trunk groups are not restricted to just two paths and more can be added to increase throughput. It is very easy to understand and to configure switches for trunking. Recovery time from a cable fault is extremely fast as the switches divert traffic to functioning paths. However, there will be a reduction in throughput until the cable fault condition is corrected.

The disadvantage of using trunking is that it requires the installation of additional cable. Depending upon the size of the trunk group, cable requirements can double or could force the purchase of larger switches. Trunking schemes are not always standardized among vendors which may restrict the purchase of all equipment from the same vendor. Although trunking supports star topology, implementing star topology becomes more involved because of the increased cable usage.

(No part of this article may be reproduced without the written consent of the Industrial Ethernet University.)